7.4 PRECISION, RECALL, AND Fβ SCORES
Of the records classified by our model as positive, what proportion are true positives? The metric addressing this question is called precision, and is defined as follows:
![equation](https://static.packt-cdn.com/products/9781119526810/graphics/images/c7-disp-0009.png)
In the field of information retrieval (e.g. search engines) the precision metric answers the question, “What proportion of the selected items is relevant?” This metric is often paired with recall, which is just another name for sensitivity.
![equation](https://static.packt-cdn.com/products/9781119526810/graphics/images/c7-disp-0010.png)
It would be useful to combine precision and recall into a single measure. To do so, we may use Fβ scores, defined as follows:
![equation](https://static.packt-cdn.com/products/9781119526810/graphics/images/c7-disp-0011.png)
for β > 0.
- When β = 1, this is called the harmonic mean of precision and recall, which are thus equally weighted in the metric F1.
- When β > 1, Fβ weights recall higher than precision.
- When β <&...